RESEARCHARTICLE EffectsofUrbanLandscapePatternonPM …hub.hku.hk/bitstream/10722/227869/1/Content.pdf · 2016. 7. 21. · tion[13]and health riskassessment ofPM 2.5 [14],attemptingtomakeclear

RESEARCH ARTICLE

Effects of Urban Landscape Pattern on PM2.5

Pollution—A Beijing Case StudyJianshengWu1,2, Wudan Xie1*, Weifeng Li3*, Jiacheng Li1,4

1 The Key Laboratory for Environmental and Urban Sciences, School of Urban Planning and Design, PekingUniversity Shenzhen, Shenzhen, China, 2 College of Urban and Environmental Sciences, Laboratory forEarth Surface Processes, Ministry of Education, Peking University, Beijing, China, 3 Department of UrbanPlanning and Design, University of Hong Kong, Hong Kong, China, 4 Department of Urban Development andLand Policy, Urban Planning & Design Institute of Shenzhen, Shenzhen, China

*[email protected] (WX); [email protected] (WL)

AbstractPM2.5 refers to particulate matter (PM) in air that is less than 2.5μm in aerodynamic diame-

ter, which has negative effects on air quality and human health. PM2.5 is the main pollutant

source in haze occurring in Beijing, and it also has caused many problems in other cities.

Previous studies have focused mostly on the relationship between land use and air quality,

but less research has specifically explored the effects of urban landscape patterns on

PM2.5. This study considered the rapidly growing and heavily polluted Beijing, China. To

better understand the impact of urban landscape pattern on PM2.5 pollution, five landscape

metrics including PLAND, PD, ED, SHEI, and CONTAG were applied in the study. Further,

other data, such as street networks, population density, and elevation considered as factors

influencing PM2.5, were obtained through RS and GIS. By means of correlation analysis

and stepwise multiple regression, the effects of landscape pattern on PM2.5 concentration

was explored. The results showed that (1) at class-level, vegetation and water were signifi-

cant landscape components in reducing PM2.5 concentration, while cropland played a spe-

cial role in PM2.5 concentration; (2) landscape configuration (ED and PD) features at class-

level had obvious effects on particulate matter; and (3) at the landscape-level, the evenness

(SHEI) and fragmentation (CONTAG) of the whole landscape related closely with PM2.5

concentration. Results of this study could expand our understanding of the role of urban

landscape pattern on PM2.5 and provide useful information for urban planning.

IntroductionPM2.5, which refers to particulate matter (PM) in air that is less than 2.5μm in aerodynamicdiameter [1], is a key pollutant affecting human health, visibility and radiation balance [2]. Thesmall size, strong adsorption and complex constitution are major features, as it can carry heavymetals and sulfates, etc. into the respiratory tract and lungs [3]. PM2.5 mainly originates fromthe products of our daily activities, such as vehicle exhaust, marine aerosols, coal and fuel oilcombustion, burning of agricultural wastes, paved road dust, and secondary sulfates, etc. [4].

PLOSONE | DOI:10.1371/journal.pone.0142449 November 13, 2015 1 / 20

OPEN ACCESS

Citation:Wu J, Xie W, Li W, Li J (2015) Effects ofUrban Landscape Pattern on PM2.5 Pollution—ABeijing Case Study. PLoS ONE 10(11): e0142449.doi:10.1371/journal.pone.0142449

Editor: Yinping Zhang, Tsinghua University, CHINA

Received: June 22, 2015

Accepted: October 20, 2015

Published: November 13, 2015

Copyright: © 2015 Wu et al. This is an open accessarticle distributed under the terms of the CreativeCommons Attribution License, which permitsunrestricted use, distribution, and reproduction in anymedium, provided the original author and source arecredited.

Data Availability Statement: All data are fullyavailable without restriction. (1) PM2.5 concentrationdata was available on the website of BeijingEnvironmental Monitoring Center. Data are availablefrom http://www.bjmemc.com.cn. (2) Land use datawere obtained from the Center for Earth SystemScience, Tsinghua University. Data are available fromhttp://data.ess.tsinghua.edu.cn/. (3) Populationdataset was from Institute of Geographic Sciencesand Natural Resource Research, CAS. Data areavailable from http://www.resdc.cn/data.aspx?DATAID=117. (4) DEM data was derived from ASTERGDEM 1st edition (V1). Data are available from http://gdem.ersdac.jspacesystems.or.jp/. (5) Other relevantdata were within the paper.

http://crossmark.crossref.org/dialog/?doi=10.1371/journal.pone.0142449&domain=pdf

http://creativecommons.org/licenses/by/4.0/

http://creativecommons.org/licenses/by/4.0/

http://www.bjmemc.com.cn

http://data.ess.tsinghua.edu.cn/

http://www.resdc.cn/data.aspx?DATAID=117

http://www.resdc.cn/data.aspx?DATAID=117

http://gdem.ersdac.jspacesystems.or.jp/

http://gdem.ersdac.jspacesystems.or.jp/

Owing to population expansion and rapid urbanization, PM2.5 has become one of the mostserious environmental problems in many cities in China [5, 6]. Big cities, such as Beijing andShanghai, have suffered terrible hazes in recent years, causing many environmental and eco-logical problems, and also making troubles in human travel [1, 7], and is drawing more andmore attention around the world. This problem is especially urgent in China, which isexperiencing rapid urbanization and is planning to continue giving high priority to urbaniza-tion in the coming decades [8].

PM2.5 pollution is a main factor in decreasing air quality. It not only causes serious visibilityproblems, but also does great harm to human health. First, PM2.5 can impair visibility by lightscattering through its suspended particles and gases, especially its components such as primaryand secondary organics, inorganic salt, inorganic carbon, and elements [9]. Second, long-termexposure to PM2.5 can result in mortality and morbidity [10]. Previous studies have confirmedthat high PM2.5 concentration is associated with increased rates of cardiovascular and respira-tory diseases [11], and also leads to cancer, as well as premature death [12]. Last, PM2.5 canaffect Earth’s radiation balance and influence energy balance and material flow [4]. Thus, howto decrease PM2.5 concentration to reduce its adverse impact has become a hot issue amongresearchers.

Recent studies have mainly focused on source appointment [6], measurement and simula-tion [13] and health risk assessment of PM2.5 [14], attempting to make clear the chemical com-ponents and sources and then try to find out some effective measures to reduce the negativeeffects. Additionally, many studies have explored the relationship between air pollution andland use [15–17]. Besides, the changes of landscape patterns, which includes the changes ofcomposition and configuration of landscape can influence PM2.5. Furthermore, various meth-ods have been explored and applied to research the factors of PM2.5 concentrations. Regressionanalysis is a common method among them[18], especially the Land Use Regression model(LUR) [19–21], which analyzes the potential factors based on GIS. Traffic, winter heating, farmburning, and land use, etc. are found to greatly influence PM2.5. Take the ESCAPE project [22]in Europe as an example, it used LUR to explore pollutant sources, including factors such asroad networks, population density, land use, and elevation (DEM), etc. Besides, a number ofstudies also concentrate on the effects of specific land use types, such as vegetation, water etc.,on PM2.5. The relationship between vegetation and air pollutant was a common focus in exper-iment and simulation studies. Sabo et al. [23] examined the PM accumulation on leaves of 22trees and 25 shrubs in test field in Norway and Poland. The results showed that Pinus mugoand Pinus sylvestris, Taxus media and Taxus baccata, Stephanandra incisa and Betula pendulawere efficient species in capturing PM. Broad-leaved species with rough leaf surfaces are moreefficient in capturing PM than those with smooth leaf surfaces [24]. Modelling approaches toresearch the PM deposition to the urban tree canopy were conducted in Japan [25], London[26] and New York [27] et al. Green land was usual variable in land-use regression models [28,29]. As for water, winds carry billions of tons of PM from the continents to the oceans. Marineand atmospheric scientists were investigating the transport and deposition of Pm to the ocean.The flux of PM depends on many factors including the distribution of sources, the physicaland chemical properties of the PM, meteorological conditions, and the rates of removal by dryand wet deposition. Atmospheric chemical transport models, coupled with in situ observations,were improving our understanding of the temporal and spatial variability of PM deposition tooceans [30]. ESCAPE case study in Stockholm County also found that water in buffer of 500meters had negative effect on PM2.5 concentration [22]. These studies all displayed the reduc-ing influence of vegetation and water on particulate matter.

The relationship between land use and PM2.5 has been confirmed in previous studies, butthere is not enough information offered in these studies to explore the effects of urban

Effects of Landscape Pattern on PM2.5

PLOS ONE | DOI:10.1371/journal.pone.0142449 November 13, 2015 2 / 20

Funding: This work was supported by: (1) 41330747,http://www.nsfc.gov.cn/, National Science Foundationof China, JW. He had a role in study design and datacollection. (2) 201311159209, http://www.hku.hk/,University of Hong Kong, WL. He had a role in studydesign and preparation of the manuscript.

Competing Interests: The authors have declaredthat no competing interests exist.

http://www.nsfc.gov.cn/

http://www.hku.hk/

landscape patterns on PM2.5 pollution. The relationship between urban landscape patterns andair pollution is a complex patterns-process relationship. Recently, with the development of RSand GIS, research has started to focus on the influence of urban form and urban landscape onair pollution. Tang and Wang [31] demonstrated that urban form had effects on traffic-induced noise and air pollution, such as urban forms in historical area with narrower roads,complex road network led to lower noise pollution, but the greater street canyon effects resultsin higher CO concentration. Weber et al. [32] revealed that landscape metrics in differenturban structures were good indicators of PM10 without measured data, which indicated thefunction of urban landscape on air pollution. An study case of evaluating how spatial heteroge-neity of forest affect air pollution was conducted by Escobedo and Nowak [33], which also tookurban landscape of green space into account. As for the relative studies in China, there werealso some papers proving the importance of landscape patterns in green space on air quality[34]. The fragmentation of green space has been shown to affect the concentration of fine par-ticulate matter in Yichang city [35]. By means of calculating and analyzing several landscapemetrics, these studies investigate how to make a better landscape planning to reduce air pollu-tion, such as how to plan vegetation to deposit PM2.5 more efficiently [36]. The reason thatlandscape pattern influence PM2.5 concentration may depend on many factors. The landscapecould firstly change the factors, such as transportation volume and wind trace and furtherinfluence the air pollutant concentration. Moreover, different landscape pattern perhapsaffected the interaction between forest, water and particulate matter in the air.

However, compare to noise pollution [31, 32], water pollution [37–39] and other ecologicalprocesses [40], there is little research focusing on the relationship between air pollution, espe-cially fine particulate matter and urban landscape patterns [33, 34]. More attention has beenpaid to the influence of different land cover on air pollution, rather than different landscapepattern. In addition, it is hard to explain the mechanism of the urban landscape pattern on par-ticulate matter though it may have some similar process with heat island. The other possiblereasons for rare studies may include the appropriateness of method, limitation of data, etc. [41,42]. As a result, firstly capturing the quantitative relationships between landscape patterns andPM2.5 is of theoretical importance and practical for optimizing urban landscape patterns andimproving air quality in the environment, especially in China today. Furthermore, it can widenour understanding of the relationship between landscape patterns and ecological process andits effects on air pollution.

Urban landscapes are characterized by complex spatial heterogeneity, as different land-cover and land-use types have their own surface characteristics. Landscape metrics are algo-rithms that quantify specific elements [32] and spatial characteristics, including patches, clas-ses, and entire landscapes, and are usually used in urban form and urban landscape research[41]. Further, landscape metrics are also applied to investigate the influence of compositionand configuration of different land use types on biodiversity [43] and habitat [44]. Addition-ally, they can be calculated quickly and directly. Thus, it is appropriate to use landscape metricsto investigate the effect of urban landscape patterns on PM2.5 concentration.

The main objective of this paper is to examine the effects of urban landscape pattern onPM2.5 pollution. The study site is Beijing, China’s capital, which has limited green space [13]and has been experiencing serious hazes in recent years [21]. Therefore, the results of thisstudy can contribute to improving urban landscape planning and management, and can be aseffective measures for addressing air quality problems in Beijing. We used variables includingPM2.5 concentration, street networks, elevation, population and landscape metrics to buildregression models of all year and each season. Then we investigated the effects of urban land-scape patterns on PM2.5 pollution in Beijing through statistical analysis. Specifically, the studyaddressed the following questions:



1. How do urban landscape composition affect PM2.5 concentration?

2. How do urban landscape configuration affect PM2.5 concentration?

3. How do whole landscape pattern affect PM2.5 concentration?

Study areaBeijing is located in the northeast of the North China Plain (115°250-117°300E, 39°280-41°250N)with a population of 20.693 million and a total area of 16410.54 km2. Its climate belongs to acontinental monsoon climate with apparent seasons. The average temperature is 12.3°C, andannual precipitation is about 572mm [45]. We used MAM (March, April and May), JJA (June,July, and August), SON (September, October, and November), and DJF (December, January,and February) to represent spring, summer, autumn, and winter respectively. As a city with ahistory of more than 3000 years, Beijing has served as the capital for more than 850 years [46].Since the implementation of China’s reform and opening-up policy in 1978, Beijing has beenundergoing rapid urbanization. Construction land area and traffic volume has increased dra-matically, which has resulted in severe air pollution. Frequent haze events and high pollutantconcentrations in this city have attracted the attention of the world. Additionally, along withurbanization, there is less impossible to increase area, especially of green space, which can miti-gate the PM2.5 concentration [31]. So how to make use of the landscape pattern to improve airquality in limited areas is of great significance for sustainable development in Beijing.

Data andmethodsTo explore the effects of landscape patterns on PM2.5 concentration, the following steps weretaken: (1) PM2.5 concentration of 35 monitoring sites in Beijing was obtained from the websiteof the Beijing Environmental Monitoring Center; (2) relevant variables, including street net-work, population, and elevation etc. were analyzed using GIS; (3) a selection of landscape met-rics were calculated in Fragstats, including class level and landscape level, to provideinformation about features of landscape composition and configuration; (4) statistical analysis,mainly including Pearson’s correlation, stepwise multiple regression and leave-one-out crossvalidation, was applied to investigate the relationship between landscape patterns and PM2.5

concentration. Specific details were provided as described in the following.

PM2.5 measurementsRoutine monitoring data were collected at 35 air quality monitoring sites, which were availableon the website of Beijing Environmental Monitoring Center in real time. Continuous hourlyPM2.5 concentrations were measured for a whole year from 4th March 2013 to 8th March,2014. The 35 sites were divided into four categories to guarantee adequate spatial variation inmeasured concentrations, including 12 urban environmental evaluation sites, 16 suburb envi-ronmental evaluation sites, 5 traffic pollution monitoring sites, and 2 regional background con-trol sites. The distribution of all sites was illustrated as shown below (Fig 1).

Land use dataLand use data for 2010 in Beijing were obtained from the Center for Earth System Science,Tsinghua University [47], derived from Landsat TM and ETM+ with a spatial resolution of30m. While the original land cover data were classified into 20 land use categories, we simpli-fied land use data into 5 categories: construction land, vegetation, water body, bare land andcropland for further analysis in landscape metrics variables.



Fig 1. Classification and distribution of air quality monitoring sites in Beijing.

doi:10.1371/journal.pone.0142449.g001



Landscape metricsBased on the land use classification, landscape metrics of different level were calculated. Land-scape metrics have been widely used to measure landscape patterns [48], which include compo-sition and configuration [49]. According to former studies [32, 50, 51], 5 landscape metrics(Table 1) were selected to measure urban landscape patterns in Beijing based on principlesincluding (1) theoretically and practically important, (2) easily calculated, (3) interpretable,and (4) little redundancy. The selected landscape metrics were applied to reflect the size, den-sity, edge, shape of different patch types, and evenness and fragmentation of the whole land-scape, respectively. The percentage of landscape (PLAND) is a landscape composition metricmeasuring the percentage of different patch types within the landscape. Patch density (PD),edge density (ED) are landscape configuration metrics describing the spatial distribution ofpatches within the landscape. Shannon’s evenness index (SHEI) and contagion (CONTAG) atlandscape level signifies some characteristics of the whole landscape. Additionally, we adaptedbuffer radii of 100, 300, 500, 1000, 2000, 3000m and 5000m around the monitoring sites. Theabove metrics describing various patch and landscape types were calculated using Fragstats4.1as variables.

Other variablesOther variables in this study also include street networks, population, and elevation etc. Streetnetworks data were obtained through map vectorization from Google Earth combined withLandsat TM 2012. We categorized all roads into major roads and secondary roads and used thelength of specific road types as traffic variables. Major roads include ring roads, expressways,and some other important roads, while secondary roads covered the rest. In accordance withthe principle of traffic-related buffer selection described by Hoek et al. [52] We set the maxi-mum traffic-related buffer distance to 1000m. Combined with dispersion patterns, we adaptedcircular buffers with 100, 200, 300, 500, 750, and 1000 m radii around the sampling sites.

Population dataset, from Institute of Geographic Sciences and Natural Resource Research,CAS, with a spatial resolution of 0.50 (816.3m approximately), was used to represent the popu-lation distribution of Beijing. Considering the spatial resolution of the dataset, we only adaptedbuffer radii of 1000, 2000 and 5000m.

DEM data, derived from ASTER GDEM 1st edition (V1) with a spatial resolution of 30m,were obtained from Geospatial Data Cloud.

Land cover data, street networks, population density, and elevation were conducted in Arc-GIS to develop the predictor variables for further study (Table 2).

Table 1. List of the selected landscapemetrics.

Metrics (abbreviation) Description (unit) Range

Percentage of landscape(PLAND)

PLAND quantifies the proportional abundance of each patch type in the landscape (percent) 0 < PLAND �100

Patch density (PD) PD expresses number of patches on a per unit area for considered class (number per 100hectares)

PD > 0

Edge density (ED) ED reports edge length on a per unit area for considered class (meter per hectare) ED � 0

Shannon’s evenness index(SHEI)

SHEI expresses the evenness distribution of area among patch types (none) 0 � SHEI � 1

Contagion(CONTAG) Tendency of land use types to be aggregated (percent) 0 < CONTAG � 100

Sources: Fragstats documents 4.2 (2014).

doi:10.1371/journal.pone.0142449.t001



Statistical analysisAnnual and seasonal average concentrations of PM2.5 in 35 sites were calculated as dependentvariables in this study, while the independent variables are listed in Table 2. The statistical anal-ysis includes Pearson’s correlation analysis, multiple stepwise regression, and leave-one-outcross validation (LOOCV). The method in this study was similar but not identical to LURmodel. The LUR model, which was developed by Briggs et al. in 1997 [53]. It uses least squaresregression to combine monitored data with GIS-based predictor data to build a predictionmodel applicable to nom-measured locations [54]. It usually contains these main steps, includ-ing obtaining the monitoring data, generating variables, developing models, model validationand regression mapping. The most prominent characteristic of LUR may be the predictor

Table 2. Classification and description of independent variables.

Class ofvariables

Description Subclass of variables Buffer radii(meters) Variables names

Street network The length of major roads and commonroads within the buffer (m)

Mr (main roads) 100;200;300;500;750;1000 Mr_xx*

Cr (common roads) 100;200;300;500;750;1000 Cr_xx

Populationdensity

Population within the buffer(in units) Pop (population) 1000;3000;5000 Pop_xx

Elevation Elevation of the site (m) DEM (elevation) DEM

Landscapemetrics

The landscape metrics of land use withinthe buffer

Crop (cropland) PLAND 100;300;500;1000;2000;3000;5000 Crop_PLAND_xx

PD 100;300;500;1000;2000;3000;5000 Crop_PD_xx

ED 100;300;500;1000;2000;3000;5000 Crop_ED_xx

SHEI 100;300;500;1000;2000;3000;5000 Crop_SHEI_xx

CONTAG 100;300;500;1000;2000;3000;5000 Crop_CONTAG_xx

Vege (vegetation) PLAND 100;300;500;1000;2000;3000;5000 Vege_PLAND_xx

PD 100;300;500;1000;2000;3000;5000 Vege_PD_xx

ED 100;300;500;1000;2000;3000;5000 Vege_ED_xx

SHEI 100;300;500;1000;2000;3000;5000 Vege_SHEI_xx

CONTAG 100;300;500;1000;2000;3000;5000 Vege_CONTAG_xx

Wat (water body) PLAND 100;300;500;1000;2000;3000;5000 Wat_PLAND_xx

PD 100;300;500;1000;2000;3000;5000 Wat_PD_xx

ED 100;300;500;1000;2000;3000;5000 Wat_ED_xx

SHEI 100;300;500;1000;2000;3000;5000 Wat_SHEI_xx

CONTAG 100;300;500;1000;2000;3000;5000 Wat_CONTAG_xx

Cons (constructionland)

PLAND 100;300;500;1000;2000;3000;5000 Cons_PLAND_xx

PD 100;300;500;1000;2000;3000;5000 Cons_PD_xx

ED 100;300;500;1000;2000;3000;5000 Cons_ED_xx

SHEI 100;300;500;1000;2000;3000;5000 Cons_SHEI_xx

CONTAG 100;300;500;1000;2000;3000;5000 Cons_CONTAG_xx

Bare (bare land) PLAND 100;300;500;1000;2000;3000;5000 Bare_PLAND_xx

PD 100;300;500;1000;2000;3000;5000 Bare_PD_xx

ED 100;300;500;1000;2000;3000;5000 Bare_ED_xx

SHEI 100;300;500;1000;2000;3000;5000 Bare_SHEI_xx

CONTAG 100;300;500;1000;2000;3000;5000 Bare_CONTAG_xx

* xx corresponds to the circular buffer radii (in meters).




variables developed in GIS. Compared to LUR model, our study added landscape metrics asvariables in the model, not only usual variables in other models, such as land use, street net-work et al., which did not occur in previous LUR models. In short, the difference between LURmodels and our method was that we made use of landscape metrics to investigate the effect oflandscape pattern on PM2.5 concentration more than land use factors.

In the first place, in order to conduct a tentative study on the effects of landscape patternson PM2.5 concentration, Pearson correlation coefficients were applied to measure the relation-ship between them. Pearson correlation coefficient is a linear correlation coefficient used toreflect the extent of correlation between two variables. The larger the absolute value of correla-tion coefficient, the greater correlation they have [48]. This study made a Pearson correlationbetween each landscape metric and PM2.5 concentration, attempting to find the landscape met-rics which have significant correlation with PM2.5 concentration (P<0.05). Pearson correlationanalysis was performed to determine whether the landscape variables and PM2.5 concentrationwere highly correlated, which laid a foundation for further study.

Next, we tested the relationship between landscape patterns and PM2.5 concentrations bystepwise multiple regression. A multiple linear regression was performed using all variables ina stepwise selection method. The method selects a subset of the variables that have a high corre-lation with dependent variables. Only variables that made a significant contribution to theoverall model were kept (P<0.05) [55, 56]. Stepwise multiple regression identifies which vari-ables explain the greatest amount of variation in PM2.5 concentration. Before stepwise multipleregression, in order to avoid the potential of collinearity among variables belonging to thesame category and ensuring interpretability of parameters, a model-building algorithm wasused, as follows[19]: (1) Remove variables with less than five nonzero values; (2) in each sub-category, rank all variables by the absolute strength of their correlation with the measured pol-lutant and identify the highest-ranking variable; (3) remove other variables in each sub-cate-gory that are correlated (Pearson’s r> 0.6) with the highest-ranking variable; (4) enter allremaining variables into a stepwise multiple linear regression with a confidence interval 95% inSPSS; (5) remove the variables that have insignificant t-statistics (P<0.05) or are inconsistentwith a priori assumptions; (6) repeat steps 4 and 5 until there are no more variables that cancontribute less than 1% to the adjusted R2. Then stepwise multiple regressions were performedto estimate the direction and magnitude of the effect of transportation, population density, ele-vation, and landscape pattern on PM2.5 pollution. There were 5 regression models in the study,including 1 annual average model and 4 seasonal average models.

In the last step, we evaluated the regression models by leave-one-out cross validation(LOOCV) [52], where models were developed for N-1 sites (N is the total number of samplingsites) and the predicted concentrations were compared with the measured concentrations atthe left-out site. The above procedure was repeated 35 times. Then, the root mean squarederror (RMSE) was calculated to describe the validity and accuracy of the models. Generally, alower RMSE value meant more stable and accurate models.

Results

Descriptive statisticsAll 35 sites were valid samples during the study period. The annual average concentration of35 sites was 90.724μg/m3, which was 2–3 times higher than the WHO Level 1 Interim Target of35μg/m3. The maximum value of annual average concentration was 115.894μg/m3 in the Liu-lihe site, while the minimum value was 62.054μg/m3 in Miyun reservoir site. Fig 2 showed thatPM2.5 concentration in winter and autumn was larger than that in spring and summer. Theaverage concentration of four seasons was 85.349μg/m3, 79.149μg/m3, 86.887μg/m3 and



112.299μg/m3. The peak concentrations of spring, summer, autumn and winter were109.930μg/m3 (Liulihe site), 100.899μg/m3 (Daxing site), 147.738μg/m3 (Daxing site) and163.347μg/m3 (Inner YongDingMen street site). The PM2.5 concentration between differentsites and different seasons changes greatly. The spatiotemporal variation of PM2.5 concentra-tion in Beijing may be evident.

Pearson correlation analysisThe relationship between PM2.5 concentration and landscape patterns was first characterizedby Pearson correlation analysis. Pearson coefficients indicated the extent of the correlation.There were 4 landscape metrics which had a significant relationship (|r|>0.6) with PM2.5 con-centration in Table 3. Firstly, among 5 land use types, only vegetation metric (vege_P-LAND_5000) related closely with PM2.5 concentration. R values between vege_PLAND_5000and annual, spring, summer, winter average concentration were -0.701, -0.790, -0.701 and-0.623 respectively, which indicated that increase of vegetation area could decrease PM2.5 con-centration, especially in spring, summer and winter. Secondly, at class-level configuration

Fig 2. Seasonal pattern of four categories of all sites.

doi:10.1371/journal.pone.0142449.g002



metrics, only vege_ED_5000 had close correlation with PM2.5 concentration in spring (r =-0.776), summer (r = -0.776) and autumn (r = -0.612) particularly. Thirdly, at landscape level,both SHEI and CONTAG displayed a significant relationship to PM2.5 concentration, and theyall belonged to the radii buffer of 3000. In addition, all metrics showed a negative relationshipwith PM2.5 concentration except CONTAG_3000. And all metrics at class-level had negativecorrelation with PM2.5 concentration, and they all related to vegetation, which was of greatimportance in mitigating PM2.5 concentration. Furthermore, from temporal perspective, theresults of correlation in summer was more similar to that in annual average. They both had 4metrics (vege_PLAND_5000, vege_ED_5000, SHEI_3000 and CONTAG_3000). While therewere only 2 metrics in spring (vege_PLAND_5000 and vege_ED_5000), and was only 1 metricin autumn (vege_ED_5000) and winter (vege_PLAND_5000). From the correlation analysis,the metrics PLAND and ED of vegetation had significant relationship with PM2.5 concentra-tion, which enhanced the importance of vegetation conservation. The SHEI and CONTAGalso signified the effects of evenness on mitigating PM2.5 concentration. They laid a foundationfor the further regression analysis.

Stepwise multiple regressionTo further analyze the contributions of landscape metrics to PM2.5 pollution in various seasons,stepwise multiple regression was employed and 5 regression models obtained. More details areshown in Table 4.

First, stepwise multiple linear regression of variables including street network, population,DEM and landscape metrics was performed to research the influence of urban landscape onPM2.5 concentration in the whole year. After removal of the non-significant variables, 6 signifi-cant variables (vege_ED_5000, crop_PLAND_1000, cons_PLAND_300, cons_ED_2000,wat_ED_3000 and mr_1000) were employed to correlate with PM2.5 concentration (Table 4).The obtained relationship is expressed by the following equation:

PM2:5year¼90:962� 0:428 � vege ED 5000 þ 0:347 � crop PLAND 1000 þ 0:125 � cons PLAND 300

�1:604 � cons PD 2000 � 0:208 � wat ED 3000 þ 0:002 �mr 1000ð1Þ

There existed negative linear correlation between PM2.5 concentration and vege_ED_5000,cons_PD_2000 and wat_ED_3000. A positive linear correlation of PM2.5 concentrationbetween crop_PLAND_1000, cons_PLAND_300 and mr_1000 was found. The order of

Table 3. Landscapemetrics that had relationship with PM2.5 concentration (|r|>0.6).

Class Class-level composition metrics (rvalue)

Class-level configuration metrics (rvalue)

Landscape-level metrics (rvalue)

Landscape metrics (annualaverage)

Vege_PLAND_5000(-0.701) Vege_ED_5000(-0.766) SHEI_3000(-0.654)

CONTAG_3000(0.631)

Landscape metrics (springaverage)

Vege_PLAND_5000(-0.790) Vege_ED_5000(-0.766)

Landscape metrics (summeraverage)

Vege_PLAND_5000(-0.701) Vege_ED_5000(-0.766) SHEI_3000(-0.654)

CONTAG_3000(0.631)

Landscape metrics (autumnaverage)

Vege_ED_5000(-0.612)

Landscape metrics (winteraverage)

Vege_PLAND_5000(-0.623)




absolute coefficients suggesting that the correlations for cons_PD_2000 was closer than other 5variables. The order of sequence into the regression model indicated that vege_ED_5000 wasof most importance for decreasing PM2.5 concentration.

Next, according to different kinds of independent variables in different period, 4 modelswere obtained as follows:

PM2:5spring ¼ 90:767� 0:485 � vege PLAND 5000 � 0:464 � bare ED 500 � 0:530

� wat PLAND 5000 ð2Þ

PM2:5summer ¼ 84:803� 0:434 � vege PLAND 5000 � 0:698 � vege PD 5000 þ 0:005 �mr 1000

�1:151 � wat PLAND 500ð3Þ

PM2:5autumn ¼ 90:819� 0:289 � vege ED 5000 þ 0:528 � crop PLAND 1000 � 0:853 � cons PD 300

�2:901 � vege PD 5000 þ 6:210 � bare PLAND 500ð4Þ

Table 4. Analysis of coefficient of regression models.

Regression model Variables Parameters of models

B t Sig.

Year Constant 90.962 14.956 0.000 Adjusted R2 = 0.849

Vege_ED_5000 -0.428 -9.472 0.000 D-W value = 2.053

Crop_PLAND_1000 0.347 5.402 0.000 RMSE = 4.754μg/m3

Cons_PLAND_300 0.125 2.629 0.014 F = 32.819(Sig. = 0.000)

Cons_PD_2000 -1.604 -3.094 0.004

Wat_ED_3000 -0.208 -2.966 0.006

Mr_1000 0.002 2.065 0.048

Spring Constant 90.767 63.759 0.000 Adjusted R2 = 0.802

Vege_PLAND_5000 -0.485 -7.572 0.000 D-W value = 1.889

Bare_ED_500 -0.464 -5.143 0.000 RMSE = 5.050μg/m3

Wat_PLAND_5000 -0.530 -4.173 0.000 F = 46.933(Sig. = 0.000)

Summer Constant 84.803 37.515 0.000 Adjusted R2 = 0.684


Vege_PD_5000 -0.698 -2.589 0.015 RMSE = 6.027μg/m3

Mr_1000 0.005 3.788 0.001 F = 19.409(Sig. = 0.000)

Wat_PLAND_500 -1.151 -3.207 0.003

Autumn Constant 90.819 24.020 0.000 Adjusted R2 = 0.624

Vege_ED_5000 -0.289 -2.521 0.017 D-W value = 2.148

Crop_PLAND_1000 0.528 5.010 0.000 RMSE = 10.317μg/m3

Cons_PD_300 -0.853 -3.190 0.003 F = 12.288(Sig. = 0.000)

Vege_PD_5000 -2.901 -3.214 0.003

Bare_PLAND_500 6.210 2.304 0.029

Winter Constant 45.596 2.288 0.029 Adjusted R2 = 0.658


Wat_ED_3000 -0.634 -4.513 0.000 RMSE = 12.956μg/m3

CONTAG_3000 1.415 4.407 0.000 F = 22.805(Sig. = 0.000)




PM2:5winter ¼ 45:596� 0:873 � vege PLAND 5000 � 0:634 � wat ED 3000 þ 1:415

� CONTAG 3000 ð5Þ

There were 3 variables appearing in the regression model of spring. From the perspective ofcoefficient, the order of the three variables was vege_PLAND_5000>bare_ED_500>wat_P-LAND_5000, showing that the composition of vegetation and water had significant relationshipwith PM2.5 concentration, and the edge density of bare land also negatively related to PM2.5 con-centration. In regression model of summer, 3 variables were landscape metrics among all 4 vari-ables except mr_1000. The maximum absolute coefficient existed in wat_PD_500, which meansit played the most related role in PM2.5 concentration. Additionally, there existed negative linearcorrelation between PM2.5 concentration and vege_PLAND_5000, vege_PD_5000 and wat_P-LAND_500, with positive linear correlation of PM2.5 concentration and mr_1000. There were 5variables in autumn model, bare land and cropland could increase PM2.5 in autumn, which con-tributed more in this season. In winter, the model showed what differently comparing to otherseasons was CONTAG_3000, the only 1 landscape metrics at landscape level, indicating the fea-ture of whole landscape also influence the particulate matter. Table 4 shows that the significanceof regression coefficients t value and models F value were less than 0.05, indicating that eachpartial regression coefficient in the regression equation were significant, and each regressionmodel was valid. The adjusted R2 of these 5 regression equations was 0.849, 0.802, 0.684, 0.624and 0.658 respectively. In the LOOCV, the RMSE for PM2.5 concentrations models were4.754μg/m3, 5.050μg/m3, 6.027μg/m3, 10.317μg/m3 and 12.956μg/m3. The results was acceptablecomparing to other studies [20, 22], especially the first 3 models.

On the basis of the above results, the classification of variables in each regression equationwas made in Table 5, according to the different variable types. As shown in Table 5, transporta-tion and landscape metrics were 2 kinds of main variables that entered into the stepwise multi-ple regressions. There were 9 class-level composition metrics in the overall models. 3 metricsbelonged to vegetation, 2 metrics belonged to water body and cropland, and the rest was

Table 5. Classification of independent variables included in regression models.

Classification Class-level compositionmetrics

Class-level configurationmetrics

Landscape-levelmetrics

Othervariables

Model year Crop_PLAND_1000(+) Vege_ED_5000(-) Mr_1000(+)

Cons_PLAND_300(+) Cons_PD_2000(-)

Wat_ED_3000(-)

Model spring Vege_PLAND_5000(-) Bare_ED_500(-)

Wat_PLAND_5000(-)

Model summer Vege_PLAND_5000(-) Vege_PD_5000(-) Mr_1000(+)

Wat_PLAND_500(-)

Model autumn Crop_PLAND_1000(+) Vege_ED_5000(-)

Bare_PLAND_500(+) Cons_PD_300(-)

Vege_PD_5000(-)

Model winter Vege_PLAND_5000(-) Wat_ED_3000(-) CONTAG_3000(+)

Land use types (number of appearing inmodels)

Vegetation (3) Vegetation (4)

Water body (2) Water body (2)

Cropland (2) Construction land (2)

Construction land(1) Bare land (1)

Bare land (1)




construction land and bare land. Class-level configuration metrics contained 9 metrics, includ-ing 4 vegetation metrics, 2 water body metrics, 2 construction land and 1 bare land metrics.Among these 5 models, only 1 landscape-level metric, CONTAG_3000 appearing in the modelof winter. Besides, other dependent variables only contained transportation indicator, mr_1000.As for their effects on PM2.5 concentration, all configuration metrics at class-level had negativerelationship with PM2.5, while CONTAG_3000 and mr_1000 related positively with PM2.5.However, the effects of composition metrics at class-level on PM2.5 varied with the landscapetypes. The composition of vegetation and water body had negative effect on PM2.5, but that ofconstruction land, bare land and cropland had opposite influence. Furthermore, the variablesalso changed with the season, the most obvious one was crop_PLAND_1000 in model autumn,which indicated the crop was a special landscape types for air pollution. In conclusion, regres-sion analysis made a further and more accurate results than correlation analysis. Different landuse had different influence on PM2.5 concentration. Vegetation and water could deposit particu-late matter, while bare land and construction land could produce particulate matter, crop landhad uncertain relationship with PM2.5 concentration. As for landscape configuration metrics,both ED and PDmay decrease PM2.5 concentration no matter what kind of land use. The even-ness of whole landscape and main road contributed the PM2.5 concentration, too.

Discussion

Effects of urban landscape composition on PM2.5 concentrationThe concentration of PM2.5 was controlled by multiple factors, such as wind, precipitation,traffic conditions etc., but this paper took major focus on urban landscape pattern. PLAND,the composition metric, characterizes the percentage of patch classification in the whole land-scape [50], from which we could learn the effect of land use types on PM2.5 indirectly. Moreinformation can be gained in regression models. In 5 models, all land use types all entered. Theland use types most frequently associated with PM2.5 concentration was vegetation, next werewater body and cropland, bare land and construction land were least. The coefficients ofvege_PLAND in model spring, model summer and model winter were -0.485, -0.434 and-0.873 respectively. The coefficients of wat_PLAND in model spring and model summer were-0.530 and -1.151 respectively. They were negative, indicating that increasing the sink land-scape percentage would decrease PM2.5 concentration. Cons_PLAND (coefficient was 0.125 inmodel year), bare_PLAND (coefficient was 6.210 in model autumn) and crop_PLAND (coeffi-cients were 0.347 and 0.528 in model year and model autumn) were proved to influence PM2.5

oppositely, for the coefficients were positive in models. In fact, it was easy to understand therole of PLAND on PM2.5. Sink landscape can absorb PM2.5, and construction land can produceparticulate matter [57]. Altering their areas caused increasing and decreasing PM2.5, demon-strating results similar to many relative studies.

A number of previous studies have shown a strong relationship with land use, mainly apply-ing LURmodels [5, 22] and simple linear regression [18]. Those studies similarly showed thatvegetation [27, 47] and water [22] were able to reduce PM2.5 concentration. As a sink landscapefor PM2.5, vegetation and water play a primary role in PM2.5 pollution, as was derived from cor-relation analysis and stepwise multiple regression. Vegetation mainly absorbed particulate mat-ter through leaves by dry and wet deposition to reduce ambient PM2.5 concentration. Treeplanning has been put forward by the Beijing municipal government as a major measure toimprove air quality. Yang et al. [58] used an urban forest effects model to explore the effect ofurban forest on air pollution. Results showed that trees in central Beijing removed 1261.4 tonsof pollutants, most of which were particulate matter. Research in 10 U.S. cities also showed simi-lar results in that the amount of PM2.5 removed by trees ranged from 4.7 tons to 64.5 tons



annually, for instance saving the state of New York 60 million dollars in healthcare costs andmortalities because of cleaner air [27]. Besides regression models, some studies based on physi-cal model also showed the similar results. Gromke [59] indicated that the trees had bad influ-ence on pollutant dispersion using a new vegetation model. Ji and Zhao [60] used drift fluxmodel and Reynolds-Average Navier-Stokes (RANS) model to investigate the effect of trees onparticle concentration distributions and building. Dzierzanowski et al. [61] further focused onthe various functions of different tree species on particulate matter. These results of above stud-ies all revealed that vegetation had the great impact on PM2.5 mitigation, which agreed with theresults of this study. Further, water played a significant role as well, primarily depositing andabsorbing particulate matter across vast surfaces, which complied with results in experimentand simulation studies [22, 62]. However, water’s effect may not obviously comparing withother land use types. So it has often been often neglected in LUR models [52]. Nonetheless, thisstudy confirmed the significance of water, which should be given more attention in future study.

Besides water body and vegetation, cropland, construction land and bare land were alsoimportant land use type in regression models. Cropland was a special factor. On the one hand,as a part of vegetation [63], it can reduce PM2.5 concentration by deposition; on the otherhand, smoke can be produced by straw burning during harvest, the reason that many citiesreach peak PM2.5 concentration in autumn, which could be shown obviously in model autumn.In addition, cropland can be considered bare land when it is left aside. Thus, the effect of crop-land on PM2.5 concentration depended on the balance between these two functions. So only inmodel autumn could crop_PLAND entered, which not only verified but highlighted the impor-tance of cropland for air pollution, in autumn particularly. Meantime, construction land shareda mass of impervious surfaces. The rapid development of urbanization has led to increasingareas of construction land, bringing burgeoning population and growing transportation use[64], increasing the potential for PM2.5. According to Tan et al. [65], who took Taiwan as a casestudy for analyzing the holiday effect on air quality, low urbanization areas always had betterair than cities in Taiwan, no matter what period of the year. As a consequence, during the pro-cess of urbanization, how to manipulate PM2.5 will be the most important issue to deal with.

Furthermore, there were some differences among four seasons. Firstly, among the four sea-sons, only autumn model contained the crop variable. Land use did not change greatly duringthe year, but crop was a special land use. In autumn, the crop could produce smoke by strawburning, either in Beijing or surrounding regions, which make it significant in autumn modelrather than other three seasons. Secondly, the winter model only contained two variables, pro-ducing a lower R2. It may be explained by specific PM2.5 sources that were distributed in a smallscale, such as meteorological conditions, fossil fuel combustion, biomass burning for cookingand winter heating [17, 66–68], and setting off fireworks in the winter [69, 70]. Thirdly, bareland could be a source because of the soil or sand dust caused by wind erosion, especially in thespring of Beijing, when sand storm happened frequently. In conclusion, due to the regressionalgorithm, which only picked up the variables that were significant, and complicated influenceof many other factors, it was not easy to explain all different variables between seasons.

Effects of urban landscape configuration on PM2.5 concentrationThe relationship between urban landscape patterns and PM2.5 concentration were not clearlynoted in previous studies. The conclusion that landscape patterns correlated to PM2.5 con-centration was obtained in our study through correlation analysis and stepwise multipleregression. Pearson correlation analysis (Table 3) showed that ED had a significant relationto PM2.5 concentration. Pearson coefficients between Vege_ED and PM2.5 concentrationwere -0.766 and -0.612 in different models. While in regression models, ED and PD were



entered as well. Results revealed ED and PD had more evident relationship with PM2.5,which was rarely referenced in other studies.

In theory, according to the relationship between PLAND and PM2.5 concentration, the prob-lem of PM2.5 could be dealt with by changing the percentage of specific patches, but there wouldbe little possibility to increase sink landscape area or decrease construction land area in manyquickly developing cities, especially in Beijing [46]. So the positive role of landscape configurationmust be given full play in solving serious haze problems. Unfortunately, few studies have investi-gated the effect of configuration on PM2.5. Morani et al. [36] only discussed best tree plantinglocations to reduce the concentration of particulate matter. Zhang et al. [71] simulated the airflowand PM2.5 dispersion making use of large eddy simulation method, which took street canyon as aspecial land structure factor. The studies of Ji and Zhao [60] also provided suggestions on effec-tive configuration for trees to reduce PM concentration. These regression and simulation resultsrevealed that the configuration may related with pollutant to some extent. Correlation analysisand stepwise multiple regression results in our study showed that not only composition but alsoconfiguration affects air pollution. In other words, we can ease haze by transforming patch per-centage and optimizing the configuration. It is noteworthy that numerous studies have focusedon the relationship between landscape patterns and heat island. In fact, the haze island caused byPM2.5 was similar to heat island to some extent. Connors et al. [72] pointed out that the impactof configuration on urban heat island was context-dependent, and the most important metricsinfluencing it were LSI and ED. Buyantuyev andWu [73] explored heat island and landscape het-erogeneity. They learned that the interaction between land utilization and patterns of humanactivities will affect city temperatures. The relationship between configuration of various patchesof urban landscape and PM2.5 concentrations can be explained by the following principles.

Both ED (edge density) and PD (patch density) can be used to represent the complexity of theedge and reflect the degree of interaction between certain landscape and ambient landscapes,Vege_ED, wat_ED, bare_ED, cons_PD, cons_PD and vege_PD in models all appeared to beimportant for explaining variation in PM2.5 concentrations. Increasing ED and PD of these landuse types can mitigate PM2.5 more efficiently based on our study. It could be explained by the fol-lowing reasons. Firstly, along with increasing ED and PD, the interaction between land use typescan be more intensive, playing a more useful role for vegetation in PM2.5 deposition. Vegetationcan absorb more particulate matter produced from other source landscapes. Secondly, urbanlandscape configuration could affect ambient microclimate, such as wind, humility and tempera-ture et al. Canyon effect was a focus researching the relationship between urban structure withclimate and human activity [32]. The street structure could change wind speed and direction.Heat island was largely influenced by urban configuration, indicating the effect of urban land-scape pattern on temperature [74]. These methodology conditions were the main factors influ-encing PM2.5. In result, ED and PD could affect the ambient air pollution in an indirect way.Lastly, the edge and patch density may change the human activity, which may be another reasonfor mitigating PM2.5 concentration. However, the increase of edge density and patch density mayenhance energy flow and exchange between green land and surrounding patches, thereforedepositing more PM2.5 from its ambient area, which leads to reducing the PM2.5 concentration[75]. Edge and patch density availability can optimize the configuration of sink landscapes, creat-ing more opportunities for source landscape and sink landscape to interact, enhancing particulatepollution removal. This is of great significance for urban landscape planning and management.

Effects of whole landscape patterns on PM2.5 concentrationThe results from our study also revealed that landscape-level metrics can affect PM2.5 concen-tration as well as class-level metrics. We chose SHEI and CONTAG to reflect the evenness and



fragmentation of the whole landscape. SHEI had significant negative correlation with PM2.5

concentration. In correlation analysis, r value between SHEI_3000 and PM2.5 concentrationwas -0.654, while the CONTAG_3000 (r = 0.631) showed a positive relation to PM2.5 concen-tration. In regression models, CONTAG_3000 was shown to effect PM2.5 concentration in win-ter. Therefore, at the landscape level, SHEI and CONTAG were two factors influencing PM2.5.

SHEI can reflect the landscape heterogeneity of patches types, and it is sensitive to the distri-bution of various patches. The larger index suggests the more well-distributed landscape. Themore well-distributed the landscape, the closer the relationship between each land use andmore interaction between “sink” landscape and “source” landscape have, thus more frequentlymitigating the PM2.5 pollution [46]. CONTAG describes the tendency of land use types to beaggregated. The smaller index means more scattered landscape, which indicates there aremany different small patches. In other words, the communication between them can be moreeffective [35]. On the other hand, humility and heat also varies with the whole landscape pat-tern. This theory was the same as that relating to ED and PD. As a consequence, we should tryto evenly distribute all kinds of patches in the whole landscape and balance the source land-scape functions and sink landscape as much as possible. The findings from our study that bothclass-level and landscape-level metrics influenced PM2.5 concentration is of significance tourban landscape planning and management.

Limitations and recommendations for future studiesThere were data from only 35 sites analyzed in this study due to number of monitoring sites inBeijing, which could reduce the precision of regression equations to a certain extent. The idealnumber of sites is 40–80, according to Hoek et al. [52]. Moreover, air pollution data was influ-enced by time and location, so it was difficult to assess the timeliness and stability of the results.It was no doubt that PM2.5 was controlled by many factors besides the variables in our study.Xie et al. [76] found that PM2.5 concentration had relationship with SO2, NO2, CO and O3

according to the case studies in 31 Chinese cities. Wu et al. [17] added canteen amount as anindependent variable in Beijing study. The recent research took meteorological factors intoaccount, such as humidity, wind speed and wind direction [77,78]. Tang et al. [79] consideredthe influence of street pattern and building height. With the rapid development of techniqueand data sharing around the world, there would be more potential variables adding to the mod-els. The selection of variables in our study partially due to the inaccessibility of more data. Fur-thermore, both adjusted R2 and RMSE, which usually used to describe the performance ofmodel, were among reasonable range comparing to others regression models [52]. So theresults in this study could be also useful for estimating the effect of landscape features on PM2.5

concentration though without considering other factors.The results from our study verified that urban landscape pattern could also affect PM2.5

concentration, which may be of some benefit for air pollution management and landscapeplanning. However, the mechanisms and processes responsible for the effects of landscape met-rics on PM2.5 pollution and seasonal differences could not be identified clearly from the statisti-cal models applied in this study. In addition, it may be difficult to expect the performance ofthe models adding more variables, but such experiment would definitely be in our furtherstudy. Other more research, such as spatiotemporal characteristics of the effects, the impactscale and intensity, and mechanisms of seasonal differences were also included.

ConclusionsThere is no doubt that PM2.5 has become a serious air pollution problem in many rapidlydeveloping cities. PM2.5 not only harms the environment, but also harms human health. So



how to mitigate PM2.5 concentration is a focus of much research. Taking Beijing, one of themost air-polluted cities in China, as an example, our study quantitatively investigated theeffects of urban landscape patterns on PM2.5 concentration using correlation analysis and step-wise multiple regression. Our study yielded the following conclusions:

1. Among landscape composition, as sink landscape for PM2.5, vegetation and water body hadthe most obvious influence on PM2.5 concentration. Vegetation and water can absorb par-ticulate matter to reduce pollutants, while construction land and bare land will increase theconcentration, and cropland was a special factor for its different function varying withseasons.

2. Configuration metrics at class-level ED and PD were significantly related with PM2.5 con-centration. The larger ED and PD of landscape would remove more PM2.5.

3. Landscape-level metrics influenced PM2.5 concentration as well as class-level metrics. Theevenness (SHEI) and fragment (CONTAG) of the whole landscape had a significant rela-tionship with PM2.5 pollution. More even and scattered landscape distribution may be betterfor mitigating the particulate matter.

The above results can provide additional useful information for better urban landscapeplanning and management.

AcknowledgmentsWe appreciate the extensive help provided by Fei Yao, helped with data processing, and Shi-quan Zhao advised on data analyses.

Author ContributionsConceived and designed the experiments: JWWXWL JL. Performed the experiments: JWWX. Analyzed the data: JWWXWL JL. Contributed reagents/materials/analysis tools: JWWXWL JL. Wrote the paper: JWWX. Paper revision and language correction: JWWXWL JL.

References1. Wang JL, Zhang YH, Shao M, Liu XL, Zeng LM, Cheng CL, et al. Quantitative relationship between visi-

bility and mass concentration of PM2.5 in Beijing. J Environ Sci (China). 2006; 18(3):475–481.

2. Sun Y, Wang Y, Zhang C. Vertical observations and analysis of PM2.5, O3, and NOx at Beijing andTianjin from towers during summer and autumn 2006. Adv Atmos Sci. 2010; 27(1):123–136.

3. Zhang Z, Tao J, Xie S, Zhou L, Song D, Zhang P, et al. Seasonal variations and source apportionmentof PM2.5 at urban area of Chengdu. Acta Scientiae Circumstantiae. 2013; 33(11):2947–2952.

4. Pui D, Chen S, Zuo Z. PM2.5 in China: Measurements, sources, visibility and health effects, and mitiga-tion. Particuology. 2014; 13:1–26.

5. Cao G, Zhang X, Gong S, An X, Wang Y. Emission inventories of primary particles and pollutant gasesfor China. Chinese Sci Bull. 2011; 56(8):781–788.

6. Li L, WangW, Feng J, Zhang D, Li H, Gu Z, et al. Composition, source, mass closure of PM2.5 aerosolsfor four forests in eastern China. J Environ Sci. 2010; 22(3):405–412.

7. Ding A, Fu C, Yang X, Sun J, Zheng L, Xie Y, et al. Ozone and fine particle in the western YangtzeRiver Delta: an overview of 1 yr data at the SORPES station. Atmos Chem Phys. 2013; 13(11):5813–5830.

8. Chen J, Li F, Xuan C. A preliminary analysis of the use of resources in intelligent information accessresearch. Proceedings of the American Society for Information Science and Technology. 2007; 43(1):1–15.

9. Sun Y, Zhuang G, Tang A, Wang Y, An Z. Chemical characteristics of PM2.5 and PM10 in haze—fog epi-sodes in Beijing. Environ Sci Technol. 2006; 40(10):3148–3155. PMID: 16749674



http://www.ncbi.nlm.nih.gov/pubmed/16749674

10. Dockery DW. Health effects of particulate air pollution. Ann Epidemiol. 2009; 19(4):257–263. doi: 10.1016/j.annepidem.2009.01.018 PMID: 19344865

11. Pope CA, Dockery DW. Health effects of fine particulate air pollution: lines that connect. J Air WasteManage. 2006; 56(6):709–742.

12. Pope CA, Burnett RT, Krewski D, Jerrett M, Shi Y, Calle EE, et al. Cardiovascular mortality and expo-sure to airborne fine particulate matter and cigarette smoke: shape of the exposure-response relation-ship. Circulation. 2009; 120(11):941–948. doi: 10.1161/CIRCULATIONAHA.109.857888 PMID:19720932

13. Wang H, Zhuang Y, Wang Y, Sun Y, Yuan H, Zhuang G, et al. Long-termmonitoring and source appor-tionment of PM2.5/PM10 in Beijing, China. J Environ Sci (China) 2008; 20(11):1323–1327.

14. Wu S, Deng F, Wang X, Wei H, Shima M, Huang J, et al. Association of lung function in a panel ofyoung healthy adults with various chemical components of ambient fine particulate air pollution in Bei-jing, China. Atmos Environ. 2013; 77:873–884.

15. Kashima S, Yorifuji T, Tsuda T, Doi H. Application of land use regression to regulatory air quality data inJapan. Sci Total Environ. 2009; 407(8):3055–3062. doi: 10.1016/j.scitotenv.2008.12.038 PMID:19185904

16. Gulliver J, Morris C, Lee K, Vienneau D, Briggs D, Hansell A. Land use regression modeling to estimatehistoric (1962–1991) concentrations of black smoke and sulfur dioxide for Great Britain. Environ SciTechnol. 2011; 45(8):3526–3532. doi: 10.1021/es103821y PMID: 21446726

17. Wu J, Li J, Peng J, Li W, Xu G, Dong C. Applying land use regression model to estimate spatial variationof PM2.5 in Beijing, China. Environ Sci Pollut R 2015; 22(9):7045–7061.

18. Rosenlund M, Forastiere F, Stafoggia M, Porta D, Perucci M, Ranzi A, et al. Comparison of regressionmodels with land-use and emissions data to predict the spatial distribution of traffic-related air pollutionin Rome. J Expo Sci Env Epid 2007; 18(2):192–199.

19. Henderson SB, Beckerman B, Jerrett M, Brauer M. Application of land use regression to estimate long-term concentrations of traffic-related nitrogen oxides and fine particulate matter. Environ Sci Technol.2007; 41(7):2422–2428. PMID: 17438795

20. Ross Z, Jerrett M, Ito K, Tempalski B, Thurston G. A land use regression for predicting fine particulatematter concentrations in the New York City region. Atmos Environ. 2007; 41(11):2255–2269.

21. Yu M, Carmichael GR, Zhu T, Cheng Y. Sensitivity of predicted pollutant levels to anthropogenic heatemissions in Beijing. Atmos Environ. 2014; 89:169–178.

22. Eeftens M, Beelen R, de Hoogh K, Bellander T, Cesaroni G, Cirach M, et al. Development of land useregression models for PM2.5, PM2.5 absorbance, PM10 and PMcoarse in 20 European study areas;results of the ESCAPE project. Environ Sci Technol. 2012; 46(20):11195–11205. doi: 10.1021/es301948k PMID: 22963366

23. Sæbø A, Popek R, Nawrot B, Hanslin HM, Gawronska H, Gawronski SW. Plant species differences inparticulate matter accumulation on leaf surfaces. Sci Total Environ. 2012; 427–428:347–354. doi: 10.1016/j.scitotenv.2012.03.084 PMID: 22554531

24. Hwang H, Yook S, Ahn K. Experimental investigation of submicron and ultrafine soot particle removalby tree leaves. Atmos Environ. 2011; 45(38):6987–6994.

25. Matsuda K, Fujimura Y, Hayashi K, Takahashi A, Nakaya K. Deposition velocity of PM2.5 sulfate in thesummer above a deciduous forest in central Japan. Atmos Environ. 2010; 44(36):4582–4587.

26. Tallis M, Taylor G, Sinnett D, Freer-Smith P. Estimating the removal of atmospheric particulate pollutionby the urban tree canopy of London, under current and future environments. Landscape Urban Plan.2011; 103(2):129–138.

27. Nowak DJ, Hirabayashi S, Bodine A, Hoehn R. Modeled PM2.5 removal by trees in ten U.S. cities andassociated health effects. Environ Pollut. 2013; 178:395–402. doi: 10.1016/j.envpol.2013.03.050PMID: 23624337

28. Lee J, Wu C, Hoek G, de Hoogh K, Beelen R, Brunekreef B, et al. LURmodels for particulate matters inthe Taipei metropolis with high densities of roads and strong activities of industry, commerce and con-struction. Sci Total Environ. 2015; 514:178–184. doi: 10.1016/j.scitotenv.2015.01.091 PMID:25659316

29. Dirgawati M, Barnes R, Wheeler AJ, Arnold A, McCaul KA, Stuart AL, et al. Development of land useregression models for predicting exposure to NO2 and NOx in metropolitan Perth, Western Australia.Environ Modell Softw. 2015:1–10.

30. Prospero JM, Arimoto R. Atmospheric transport and deposition of particulate material to the oceans.Encyclopedia of Ocean Sciences ( Second Edition).2009:248–257.

31. Tang UW,Wang ZS. Influences of urban forms on traffic-induced noise and air pollution: results from amodelling system. Environ Modell Softw. 2007; 22(12):1750–1764.



http://dx.doi.org/10.1016/j.annepidem.2009.01.018

http://dx.doi.org/10.1016/j.annepidem.2009.01.018


http://dx.doi.org/10.1161/CIRCULATIONAHA.109.857888


http://dx.doi.org/10.1016/j.scitotenv.2008.12.038


http://dx.doi.org/10.1021/es103821y



http://dx.doi.org/10.1021/es301948k

http://dx.doi.org/10.1021/es301948k





http://dx.doi.org/10.1016/j.envpol.2013.03.050




32. Weber N, Haasea D, Franck U. Assessing modelled outdoor traffic-induced noise and air pollutionaround urban structures using the concept of landscapemetrics. Landscape Urban Plan. 2014;125:105–116.

33. Escobedo FJ, Nowak DJ. Spatial heterogeneity and air pollution removal by an urban forest. Land-scape Urban Plan. 2009; 90(3–4):102–110.

34. Ding Y, Li G, Lu X, Gao M. Spatial heterogeneity and air pollution removal by green space in GreaterPearl River Delta. Progress in Geography. 2011; 30(11):1415–1421.

35. Shao T, Zhou Z, Wang P, TangW, Liu X, Hu X. Relationship between urban green-land landscape pat-tern and air pollution in the central district of Yichang city. Chinese J Appl Ecol. 2004; 15(4):691–696.

36. Morani A, Nowak DJ, Hirabayashi S, Calfapietra C. How to select the best tree planting locations toenhance air pollution removal in the million trees NYC initiative. Environ Pollut. 2011; 159(5):1040–1047. doi: 10.1016/j.envpol.2010.11.022 PMID: 21168939

37. Lee S, Hwang S, Lee S, Hwang H, Sung H. Landscape ecological approach to the relationships of landuse patterns in watersheds to water quality characteristics. Landscape Urban Plan. 2009; 92(2):80–89.

38. Łowicki D. Prediction of flowing water pollution on the basis of landscape metrics as a tool supportingdelimitation of Nitrate Vulnerable Zones. Ecol Indic. 2012; 23:27–33.

39. Tu J. Spatially varying relationships between land use and water quality across an urbanization gradi-ent explored by geographically weighted regression. Appl Geogr. 2011; 31(1):376–392.

40. Schindler S, vonWehrden H, Poirazidis K, Wrbka T, Kati V. Multiscale performance of landscape met-rics as indicators of species richness of plants, insects and vertebrates. Ecol Indic. 2013; 31:41–48.

41. Schwarz N. Urban form revisited—Selecting indicators for characterising European cities. LandscapeUrban Plan. 2010; 96(1):29–47.

42. Wu J, Jenerette GD, Buyantuyev A, Redman CL. Quantifying spatiotemporal patterns of urbanization: Thecase of the two fastest growing metropolitan regions in the United States. Ecol Complex. 2011; 8(1):1–8.

43. Uuemaa E, Antrop M, Roosaare J, Marja R, Mander Ü. Landscape metrics and indices: an overview oftheir use in landscape research. Living Rev. Landscape Res. 2009; 5–28.

44. Santos-Filho M, Peres CA, Da Silva DJ, Sanaiotti TM. Habitat patch and matrix effects on small-mam-mal persistence in Amazonian forest fragments. Biodivers Conserv. 2012; 21(4):1127–1147.

45. Li X, ZhouW, Ouyang Z. Relationship between land surface temperature and spatial pattern of green-space: what are the effects of spatial resolution. Landscape Urban Plan. 2013; 114:1–8.

46. Li X, ZhouW, Ouyang Z, XuW, Zheng H. Spatial pattern of greenspace affects land surface tempera-ture: evidence from the heavily urbanized Beijing metropolitan area, China. Landscape Ecol. 2012; 27(6):887–898.

47. Gong P, Wang J, Yu L, Zhao Y, Zhao Y, Liang L, et al. Finer resolution observation and monitoring ofglobal land cover: first mapping results with Landsat TM and ETM+ data. Int J Remote Sens. 2013; 34(7):2607–2654.

48. McGarigal K, Cushman SA, Neel MC. Ene E. (2002). FRAGSTATS: Spatial pattern analysis programfor categorical maps. Computer software program produced by the authors at the University of Massa-chusetts, Amherst 2002.

49. Chen A, Sun R, Chen L. Effects of urban green pattern on urban surface thermal environment. ActaEcologica Sinica. 2013; 33(8):2372–2380.

50. Maimaitiyiming M, Ghulam A, Tiyip T, Pla F, Latorre-Carmona P, Halik Ü, et al. Effects of green spacespatial pattern on land surface temperature: Implications for sustainable urban planning and climatechange adaptation. Isprs J Photogramm. 2014; 89:59–66.

51. Shen Z, Hou X, Li W, Aini G. Relating landscape characteristics to non-point source pollution in a typicalurbanized watershed in the municipality of Beijing. Landscape Urban Plan. 2014; 123:96–107.

52. Hoek G, Beelen R, de Hoogh K, Vienneau D, Gulliver J, Fischer P, et al. A review of land-use regressionmodels to assess spatial variation of outdoor air pollution. Atmos Environ. 2008; 42(33):7561–7578.

53. Briggs DJ, Collins S, Elliott P, Fischer P, Kingham S, Lebret E, et al. Mapping urban air pollution usingGIS: a regression-based approach. Int J Geogr Inf Sci. 1997; 11(7):699–718.

54. de Hoogh K, Korek M, Vienneau D, Keuken M, Kukkonen J, NieuwenhuijsenMJ, et al. Comparing landuse regression and dispersionmodelling to assess residential exposure to ambient air pollution for epide-miological studies. Environ Int. 2014; 73:382–392. doi: 10.1016/j.envint.2014.08.011 PMID: 25233102

55. Howell CA, Latta SC, Donovan TM, Porneluzi PA, Parks GR, Faaborg J. Landscape effects mediatebreeding bird abundance in midwestern forests. Landscape Ecol. 2000; 15(6):547–562.

56. Saveraid EH, Debinski DM, Kindscher K, Jakubauskas ME. A comparison of satellite data and land-scape variables in predicting bird species occurrences in the Greater Yellowstone Ecosystem, USA.Landscape Ecol. 2001; 16(1):71–83.





http://dx.doi.org/10.1016/j.envint.2014.08.011


57. Chen L, Peng S, Liu J, Hou Q. Dry deposition velocity of total suspended particles and meteorologicalinfluence in four locations in Guangzhou, China. J Environ Sci. 2012; 24(4):632–639.

58. Yang J, McBride J, Zhou J, Sun Z. The urban forest in Beijing and its role in air pollution reduction.Urban For Urban Gree. 2005; 3(2):65–78.

59. Gromke C. A vegetation modeling concept for building and environmental aerodynamics wind tunneltests and its application in pollutant dispersion studies. Environ Pollut. 2011; 159(8–9):2094–2099. doi:10.1016/j.envpol.2010.11.012 PMID: 21131112

60. Ji W, Zhao B. Numerical study of the effects of trees on outdoor particle concentration distributions.Build Simul-China. 2014; 7(4):417–427.

61. Dzierżanowski K, Popek R, Gawrońska H, Sæbø A, Gawroński SW. Deposition of particulate matter ofdifferent size fractions on leaf surfaces and in waxes of urban forest species, Int J of Phytoremediat.2011: 13(10), 1037–1046.

62. Boyd PW, Mackie DS, Hunter KA. Aerosol iron deposition to the surface ocean—modes of iron supplyand biological responses. Mar Chem. 2010; 120(1–4):128–143.

63. Tao Y, Li F, Wang R, Zhao D. Research progress in the quantitative methods of urban green space pat-terns. Acta Ecologica Sinica. 2013; 33(8):2330–2340.

64. Duh J, Shandas V, Chang H, George LA. Rates of urbanisation and the resiliency of air and water qual-ity. Sci Total Environ. 2008; 400(1–3):238–256. doi: 10.1016/j.scitotenv.2008.05.002 PMID: 18603283

65. Tan P, Chou C, Chou CCK. Impact of urbanization on the air pollution “holiday effect” in Taiwan. AtmosEnviron. 2013; 70:361–375.

66. Carter EM, Shan M, Yang XD, Li JR, Baumgartner J. Pollutant emissions and energy efficiency of Chi-nese gasifier cooking stoves and implications for future intervention studies. Environ Sci Technol.2014; 48:6461–6467 doi: 10.1021/es405723w PMID: 24784418

67. Cheng Y, Engling G, He KB, Duan FK, Ma YL, Du ZY, et al. Biomass burning contribution to Beijingaerosol. Atmos Chem Phys. 2013; 13:7765–7781.

68. Yu L, Wang G, Zhang R, Zhang L, Song Y, Wu B, et al. Characterization and source apportionment ofPM2.5 in an urban environment in Beijing. Aerosol Air Qual Res. 2013; 13:574–583

69. Zhang M, Wang X, Chen J, Cheng T, Wang T, Yang X, et al. Physical characterization of aerosol parti-cles during the Chinese New Year’s firework events. Atmos Environ. 2010; 44:5191–5198.

70. Wang Y, Zhuang G, Xu C, An Z. The air pollution caused by the burning of fireworks during the lanternfestival in Beijing. Atmos Environ. 2007; 41:417–431.

71. Zhang Y, Gu Z, Lee S, Fu T, Ho K. Numerical simulation and in situ investigation of fine particle disper-sion in an actual deep street canyon in Hong Kong. Indoor Built Environ. 2011; 20: 206–216.

72. Connors JP, Galletti CS, ChowWTL. Landscape configuration and urban heat island effects: assessingthe relationship between landscape characteristics and land surface temperature in Phoenix, Arizona.Landscape Ecol. 2013; 28(2):271–283.

73. Buyantuyev A, Wu J. Urban heat islands and landscape heterogeneity: linking spatiotemporal varia-tions in surface temperatures to land-cover and socioeconomic patterns. Landscape Ecol. 2010; 25(1):17–33.

74. ZhouW, Qian Y, Li X, Li W, Han L. Relationships between land cover and the surface urban heat island:seasonal variability and effects of spatial and thematic resolution of land cover data on predicting landsurface temperatures. Landscape Ecol. 2014; 29(1):153–167.

75. ZhouW, Huang G, Cadenasso ML. Does spatial configuration matter? Understanding the effects ofland cover pattern on land surface temperature in urban landscapes. Landscape Urban Plan. 2011;102(1):54–63.

76. Xie Y, Zhao B, Zhang L, Luo R. Spatiotemporal variations of PM2.5 and PM10 concentrations between31 Chinese cities and their relationships with SO2, NO2, CO and O3. Particuology. 2015; 20:141–149.

77. McKendry IG. PM10 levels in the Lower Fraser Valley, British Columbia, Canada: An overview of spatio-temporal variations and meteorological controls. J Air Waste Manage Assoc. 2000; 50:443–452.

78. Abernethy RC, Allen RW, McKendry IG, Brauer M. A land use regression model for ultrafine particles inVancouver, Canada. Environ Sci Technol. 2013; 47:5217–5225. doi: 10.1021/es304495s PMID:23550900

79. Tang R, Blangiardo M, Gulliver J. Using building heights and street configuration to enhance intraurbanPM10, NO(X), and NO2 land use regression models. Environ Sci Technol. 2013; 47:11643–11650.doi: 10.1021/es402156g PMID: 24001269







http://dx.doi.org/10.1021/es405723w


http://dx.doi.org/10.1021/es304495s


http://dx.doi.org/10.1021/es402156g


RESEARCHARTICLE EffectsofUrbanLandscapePatternonPM …hub.hku.hk/bitstream/10722/227869/1/Content.pdf · 2016. 7. 21. · tion[13]and health riskassessment ofPM 2.5 [14],attemptingtomakeclear

Documents